Md Rezaul Huda Chowdhury Reza. Concurrency Control Modify concurrency control schemes for use in...

Md Rezaul Huda Chowdhury Reza

Modify concurrency control schemes for use in distributed environment.

We assume that each site participates in the execution of a commit protocol to ensure global transaction automicity.

We assume all replicas of any item are updated


System maintains a single lock manager that resides in a single chosen site, say Si

When a transaction needs to lock a data item, it sends a lock request to Si and lock manager determines whether the lock can be granted immediatelyIf yes, lock manager sends a message to the site

which initiated the requestIf no, request is delayed until it can be granted, at

which time a message is sent to the initiating siteMd Rezaul Huda Chowdhury Reza

The transaction can read the data item from any one of the sites at which a replica of the data item resides.

Writes must be performed on all replicas of a data item

Advantages of scheme:Simple implementationSimple deadlock handling

Disadvantages of scheme are:Bottleneck: lock manager site becomes a

bottleneckVulnerability: system is vulnerable to lock

manager site failure.Md Rezaul Huda Chowdhury Reza

Distributed Lock ManagerIn this approach, functionality of locking is

implemented by lock managers at each siteLock managers control access to local data items

But special protocols may be used for replicasAdvantage: work is distributed and can be

made robust to failuresDisadvantage: deadlock detection is more

complicatedLock managers cooperate for deadlock detection

Several variants of this approachPrimary copyMajority protocolBiased protocolQuorum consensus


Consider the following two transactions and history, with item X and transaction T1 at site 1, and item Y and transaction T2 at site 2:

T1: write (X)write (Y)

T2: write (Y)write (X)

X-lock on Xwrite (X)

X-lock on Ywrite (Y)

wait for X-lock on XWait for X-lock on Y

Result: deadlock which cannot be detected locally at either site


A global wait-for graph is constructed and maintained in a single site; the deadlock-detection coordinatorReal graph: Real, but unknown, state of the system.Constructed graph:Approximation generated by the

controller during the execution of its algorithm .the global wait-for graph can be constructed when:

a new edge is inserted in or removed from one of the local wait-for graphs.

a number of changes have occurred in a local wait-for graph.

the coordinator needs to invoke cycle-detection.If the coordinator finds a cycle, it selects a victim

and notifies all sites. The sites roll back the victim transaction.


AvailabilityHigh availability: time for which system is not

fully usable should be extremely low (e.g. 99.99% availability)

Failures are more likely in large distributed systems

To be robust, a distributed system must Detect failuresReconfigure the system so computation may continueRecovery/reintegration when a site or link is repaired

Failure detection: distinguishing link failure from site failure is hard (partial) solution: have multiple links, multiple link

failure is likely a site failureMd Rezaul Huda Chowdhury Reza

ReconfigurationReconfiguration:

Abort all transactions that were active at a failed site Making them wait could interfere with other transactions

since they may hold locks on other sites However, in case only some replicas of a data item failed,

it may be possible to continue transactions that had accessed data at a failed site (more on this later)

If replicated data items were at failed site, update system catalog to remove them from the list of replicas. This should be reversed when failed site recovers, but

additional care needs to be taken to bring values up to date

If a failed site was a central server for some subsystem, an election must be held to determine the new server E.g. name server, concurrency coordinator, global

deadlock detectorMd Rezaul Huda Chowdhury Reza

Reconfiguration (Cont.)Since network partition may not be

distinguishable from site failure, the following situations must be avoidedTwo or more central servers elected in distinct

partitionsMore than one partition updates a replicated data item

Updates must be able to continue even if some sites are down

Solution: majority based approachAlternative of “read one write all available” is

tantalizing but causes problems


Site ReintegrationWhen failed site recovers, it must catch

up with all updates that it missed while it was downProblem: updates may be happening to items

whose replica is stored at the site while the site is recovering

Solution 1: halt all updates on system while reintegrating a site Unacceptable disruption

Solution 2: lock all replicas of all data items at the site, update to latest version, then release locks Other solutions with better concurrency also

availableMd Rezaul Huda Chowdhury Reza

Heterogeneous Distributed DatabasesMany database applications require data from a

variety of preexisting databases located in a heterogeneous collection of hardware and software platforms

Data models may differ (hierarchical, relational , etc.)

Transaction commit protocols may be incompatibleConcurrency control may be based on different

techniques (locking, time stamping, etc.)System-level details almost certainly are totally

incompatible.A multi database system is a software layer on top

of existing database systems, which is designed to manipulate information in heterogeneous databasesCreates an illusion of logical database integration

without any physical database integration


Preservation of investment in existinghardwaresystem softwareApplications

Local autonomy and administrative control Allows use of special-purpose DBMSsStep towards a unified homogeneous DBMS

Full integration into a homogeneous DBMS faces Technical difficulties and cost of conversion Organizational/political difficulties

Organizations do not want to give up control on their data

Local databases wish to retain a great deal of autonomy


Unicast, Broadcast versus Multicast

Unicast One-to-one Destination – unique

receiver host addressBroadcast

One-to-all Destination – address of

networkMulticast

One-to-many Multicast group must be

identified Destination – address of

group

Key:

Unicast transfer

Broadcast transfer

Multicast transfer


Multicast application examplesFinancial services

Delivery of news, stock quotes, financial indices, etc

Remote conferencing/e-learningStreaming audio and video to many participants

(clients, students)Interactive communication between participants

Data distributione.g., distribute experimental data from Large Hadron

Collider (LHC) at CERN lab to interested physicists around the world


•Highly efficient bandwidth usageKey Architectural Decision: Add support for multicast in IP layer

Berkeley

Gatech Stanford

CMU

Routers with multicast support


So what is the big issue …

more than 20 years since proposal, but no wide area IP multicast deployment

Scalability (with number of groups)-- Routers maintain per-group state

IP Multicast: best-effort multi-point delivery service-- Providing higher level features such as reliability, congestion

control, flow control, and security has shown to be more difficult than in the unicast case

Can we achieve efficient multi-point delivery without IP-layer support?


Stanford

CMU

Stan1

Stan2

Berk2

Overlay Tree

Gatech

Berk1

Berkeley

Gatech Stan1

Stan2

Berk1

Berk2

CMU


Pros and ConsScalability

Routers do not maintain per-group stateEnd systems do, but they participate in very few groups

Potentially simplify support for higher level functionalityLeverage computation and storage of end systemsLeverage solutions for unicast congestion, error and flow control

Efficiency concernsredundant traffic on physical linksincrease in latency due to end-systems


Multicasting Algorithms



Requirements of Multipoint Routing Algorithms. Support reliable transmission

link failure should not increase delay or reduce resource availability.

Return optimal routes taking into consideration price to be paid (bandwidth consumed)end to end delay. (no. of links traversed)

Minimize network load.Avoid loops.Avoid traffic concentration on a few links or sub-nets.

Minimize the state stored in routers.



Multipoint Routing algos

Performance MetricsQuality of a tree is judged according to the following three dimensions

Low Delay:End to end delay between source and receiver relative

to the shortest unicast path delay.Low Cost :

Cost of total bandwidth consumptionCost of tree state info

Light Traffic Concentration :Maximum number of flows on a unidirectional link.How evenly the routes are distributed.



Routing AlgorithmsAll multi-point services use some kind of a distribution

tree.

Multicast trees can be

Shared across sources. (shared trees)Only one tree needs to be established for each group, which is

shared by all the sources within that group.

Source specific. (shortest path trees).A shortest path tree rooted at each sending node needs to be

established



SOURCE BASED MULTIPOINT ROUTING

The Technique.A Source Rooted Shortest Path Tree (SRSPT) algo:

Computes the shortest paths between the source and each of the receivers within the group.

Eliminates duplicate data copies on common links.Maintains one SRSPT per sender.

Concept: All receiving nodes compute path towards the source independently.

Used by: current day IP multicast protocols as applications are stillsmall scale. local area.



SOURCE BASED MULTIPOINT ROUTING

Merits vs DemeritsAdvantages.

SRSPTs are easy to compute. Use the classic unicast routing tables.

Efficient distributed implementations are possible Entire global topology

not required. There can be no loops in

the path returned.

Disadvantages

Does not minimize total cost of distribution

Does not scale well. One piece of state

information per source and per group is kept in each router.

May fail badly if the underlying unicast routing is asymmetric.



SHARED TREE APPROACH OF MULTIPOINT ROUTING

Characteristics of Steiner Tree based algorithms.The Minimum Steiner Tree: The minimal cost subgraph spanning a given subset of nodes in a graph.

The Steiner Tree problem is NP-complete.finding the minimum steiner tree in a graph has exponential

cost.

The tree designed is undirected.solution feasible only for symmetric links.

Monolithic algorithm.has to be run each time group membership changes.




Characteristics of Steiner Tree based algorithms.

The SMT defines an absolute limit on the minimum tree cost to serve as a reference for gauging the cost-optimality of heuristic alternatives.

The SMT for all members of a multicast group is the same irrespective of the role of sender or receiver. only one state entry needs to be maintained per group. it scales well for larger groups.

The SMT may have unbounded delay. Worst case maximum end-to-end path length of a SMT can

be the longest acyclic path within the graph.




Characteristics of Core Based Tree algorithms. Concept:

Use the shortest Path Tree rooted at a node in the center of the network

Steps:Choose an optimal center for the group. Multiple cores

can be used for better fault tolerance & delay characteristics.

Group members send a join message to the center. Intermediate nodes mark interface from which the

multicast info is received and forward it to the center. Choose the center to:

minimize max/avg delay for all members on the tree.Minimize the sum of tree-link costs.




Advantages of Core Based Tree algorithms Work well with multiple senders/receivers

state information is stored per group, therefore scalable.

Receiver based approach.Supports dynamic group membership with relative

ease. Suitable for sparsely distributed receivers.

SPTs will not have many common links. Do not have the unbounded delay problems of SMTs. Simple to implement

used as the basis of PIM and of The CBT interdomain Routing Protocol.




Disadvantages of Core Based Tree algorithms Incur extra delay as compared to the RPF approach.

Suffer from traffic concentration on links converging towards the center.

Choosing the optimal center is an NP complete problem.

Locating the center requires complete knowledge of the network topology.



MULTIPOINT ROUTING

TradeOffs between algos Any single tree cannot achieve Minimal Cost and

Minimal Delay both.Shortest Path Trees Minimize delay at expense of Cost.

Steiner Minimal Trees Minimize cost at expense of Delay.

Between these spectrum of different types of trees offering different tradeoffs.

Different strategies to place the routes results in different degrees of traffic concentration.


Md Rezaul Huda Chowdhury Reza. Concurrency Control Modify concurrency control schemes for use in...

Documents

Transcript of Md Rezaul Huda Chowdhury Reza. Concurrency Control Modify concurrency control schemes for use in...